Research On Filtering And Compression Methods Based On Point Cloud Data

At present,the research of hardware system represented by 3D laser scanner is relatively perfect,and the post-processing research of point cloud data is still insufficient.No matter what kind of equipment and how to collect point cloud data,there are more or less redundant points.In addition,the point cloud data have the characteristics of grid and mass,and the collected point cloud data are generally at the gigabyte(GB)level or even the terabyte(TB)level.Therefore,the storage and processing of point cloud data put forward high requirements for computer hardware.Therefore,it is of great significance to study the key technologies in point cloud filtering and compression in this paper and propose corresponding improvements.In this paper,the point cloud filtering methods are divided into two categories,one is statistical filtering,radius filtering,bilateral filtering,etc.;the other is the filtering method for ground objects.In view of the problems existing in the traditional mathematical morphology method,this paper puts forward some improvements: 1.Using adjacent ground points to fill the blank grid.2.Building triangulation network to determine local terrain slope.3.The setting of the ground elevation difference threshold of the original laser point cloud is improved.Through the experiments on the official data set of ISPRS and the point cloud of the test field,it is proved that the improved method can effectively reduce the class II error and the total error,and the feasibility is good.In terms of compression,this paper first conducts experiments and analysis on point cloud attribute compression based on TMC13.Then,the grid voxel compression algorithm is optimized by establishing an octree structure for the point cloud and taking the nearest neighbor of the voxel center point.Through analysis,when the table model obtained by statistical filtering in Chapter 3 is used,PSNR is increased by about 16.36%compared with the random sampling algorithm,16.61% compared with the curvature compression algorithm,and 3.19% compared with the uniform grid algorithm.When using the ground model obtained by the improved morphological filtering method in Chapter 3,PSNR is better than random sampling method and uniform grid method.Finally,this paper improves the compression method based on X-Y boundary extraction by extracting the boundary information of the model based on the angle between the tangent plane interior point and the neighborhood point projection.This paper uses the table model,the ground model after filtering in the third chapter and the bunny model proposed by Stanford University to carry out experiments.Through the triangular mesh reconstruction of the results of this method and the compression method based on X-Y boundary extraction,the number of triangles formed by the bunny model is 14287 and 10175,respectively,which proves that this method has good applicability and can better retain boundary characteristics at the same compression rate.